AMD Bristol/Stoney Ridge Thread

[Insert_Nickname]

I really don't understand the need for a weak, low power GPU card (I mean, beyond a certain point). I can understand the desire for a cheap card, but ...

The whole point of such a card would be replacing outdated IGPs and add-in cards*. Not everybody can just wholesale replace PCs every 3rd year. Desktops have potential 10+ year lifespans these days. There isn't much of a market in the developed world, but I could see such would be useful elsewhere, where 2nd hand or even 3rd hand PCs are common.

Now, if such would be worth the resources to produce is a whole other question.

*With the side benefit of cutting power use massively in some cases. Older cards without proper power management are powerhogs at idle.

 

[Abwx]

I bought an FX 8350 recently, new in box. Date on CPU is 2011, date on box is 2015.

Then It was manufactured in 2015, 2011 is the date at wich the trade mark for those CPUs was registered by AMD.

Btw the FX8350 was launched in october 2012...

 

[ET]

Not everybody can just wholesale replace PCs every 3rd year.

I think that having discrete GPUs pretty much makes sure that there's never need to replace complete PCs. Someone can stick a GeForce 1030 in after 5 years, and they'd be fine for another 5. What I don't get is the point of something even worse performing that that, and what I don't understand even more is why the desire for something that consumes less power than that.

But really, if someone wants an easy to use and cheap GPU, a USB GPU would be fine, assuming that nothing like gaming is need. If it's needed, then the entire idea of something lower cost than a 1030 is pointless.

Edit: By the way, I disagree with the 'potential 10 years lifespan'. That's true to an extent, but not necessarily true if you buy a low end PC, which is what you're alluding to. Websites today require a lot more RAM than 10 years ago, for example. It may be better to spend $200 on a new PC every 5 years (Athlon 200GE style PC).

 

[VirtualLarry]

It may be better to spend $200 on a new PC every 5 years (Athlon 200GE style PC).

If only I could convince my clients ("friends and family") that that was true.

I've got people holding on to 5-10 year old PCs. Not that they're unusable, like a P4 these days would be, they have SSDs of course, thanks to me. But it would be nice to get some business once in a while. Then again, other than the "Gaming PC" segment, most people's PCs, and system requirements, are largely stagnant. Sure, more RAM would help, as does a decent ad-blocker for the web. But it's undeniable, that web sites / pages are getting more content-heavy. Web sites that open a 1080P video just for the start of an article, etc.
Keeping up with GPU acceleration standards / hardware codec support is important for the web, too.

 

[NostaSeronx]

What would a 4 thread CCX look like in a Nosta APU?

Two Cluster-based Multithreading modules with two clusters/cores each or a single cluster-based multithreading module with four clusters/cores.

What's the advantage of smaller CU?

Lower power consumption, higher possible frequencies, with the same performance or negligible loss in performance. The Small CU has better performance per mm squared than the Standard CU.

a4-9120C => 600 MHz
22FDX w/ SmCU => 1.2 GHz
then, there is added boost.
22FDX successor of a6-9220C => 1.2 GHz that boosts up to 1.7 GHz
22FDX successor of a4-9120C => 1.2 GHz that boosts up to 1.4 GHz

DDR4-3600 with maximum DDR5 being 50% added to that..
DDR4-3600 maximum for DDR4 and DDR5-5400 maximum for DDR5.

We'll have to take 3 small CUs are better than 6 standard CUs. Specifically, for a 22FDX processor.

/22FDX Transistors
/New Compute Unit
/New Circuits/Cells for Logic+File+Cache
/IVR+AVFS improvements from FDSOI
]To get the 2x increase.
Then, the boost comes from extra stuff, like FBB/RBB.

Could a CU be used for FPU calculations? They had a whole HSA concept 5+ years ago. I think they should get back to it now that Zen is so well along and can take the backburner a bit.

Yes, but it goes through a decode process twice.

CPU core; ALU + AGU + FPU + FCU

FCU => Forwarder to Compute Unit;
The FCU's job is to make Graphic Core Next instructions from x86 FPU instructions. It is implied these x86 FPU ops would be a separate prefix than standard FPU ops.

In the more modern cases however, the FCU is integrated to the CPU front-end and the GCN back-end(TMUs/Graphic L1D/RBE) is integrated to the CPU back-end.

It is unlikely to get either options.

 

[Insert_Nickname]

I think that having discrete GPUs pretty much makes sure that there's never need to replace complete PCs. Someone can stick a GeForce 1030 in after 5 years, and they'd be fine for another 5. What I don't get is the point of something even worse performing that that, and what I don't understand even more is why the desire for something that consumes less power than that.

But really, if someone wants an easy to use and cheap GPU, a USB GPU would be fine, assuming that nothing like gaming is need. If it's needed, then the entire idea of something lower cost than a 1030 is pointless.

It's not a performance issue as such. Most casual users just browse the web, and don't play demanding games. We'd just need something with basic features and video acceleration for modern codecs.

It's actually more of a cost issue, the cheapest 1030 I could find with a quick search was $80. Cheapest GT710 is $30. It's an awful lot easier to convince someone to spend $30 on their PC for an improved experience then $80.

Edit: By the way, I disagree with the 'potential 10 years lifespan'. That's true to an extent, but not necessarily true if you buy a low end PC, which is what you're alluding to. Websites today require a lot more RAM than 10 years ago, for example. It may be better to spend $200 on a new PC every 5 years (Athlon 200GE style PC).

If only I could convince my clients ("friends and family") that that was true.

I've got people holding on to 5-10 year old PCs. Not that they're unusable, like a P4 these days would be, they have SSDs of course, thanks to me. But it would be nice to get some business once in a while. Then again, other than the "Gaming PC" segment, most people's PCs, and system requirements, are largely stagnant. Sure, more RAM would help, as does a decent ad-blocker for the web. But it's undeniable, that web sites / pages are getting more content-heavy. Web sites that open a 1080P video just for the start of an article, etc.
Keeping up with GPU acceleration standards / hardware codec support is important for the web, too.

Pretty much what VL said. Now, I completely agree with you in theory. In practice however, things work a bit differently. Mostly due to the fact there are a lot of cheapskates out there.

 

[amd6502]

It's not a performance issue as such. Most casual users just browse the web, and don't play demanding games. We'd just need something with basic features and video acceleration for modern codecs.

This is kind of why I think a Vega 3 display driver would be a more important feature in the Ryzen 3000 generation of CPUs than more cores (>8). Since 14nm is cheap in comparison and it might add ~50mm2 to the IO-hub it would be well worth it for the tiny %-cost increase.

Thats highly unlikely. That kind of design would be too costly for a budget GPU due to VRM requirments and you'd need a chipset as well as far as i'm aware. Neat idea though.

Well, Jaguar and 14h+16h carizzo-L were pretty low cost as far as VRM needs. In my opinion none of those cat APU's should have arrived as dual core APUs for consumers. And dual core APU salvage should've gone to embedded thin clients, or have had cores disabled for low end GPU. For consumer 3c/3t APU would have been the minimum for 14h, and near the minimum for 16h. For segmentation versus 4c/4t it would have been just right, especially with somewhat lower frequencies versus the well binned quadcores.

I think on my low end r7 250 card I see a 2 or 2+1 phase VRM.

 

[Abwx]

Some "new" 2C Bristol Ridge offering at the lowest end of the spectre...

https://geizhals.de/amd-a6-9400-ad9400agabbox-a1988913.html?hloc=at&hloc=de

 

[NTMBK]

Some "new" 2C Bristol Ridge offering at the lowest end of the spectre...

https://geizhals.de/amd-a6-9400-ad9400agabbox-a1988913.html?hloc=at&hloc=de

Oh the excitement.

 

[hojnikb]

This is kind of why I think a Vega 3 display driver would be a more important feature in the Ryzen 3000 generation of CPUs than more cores (>8). Since 14nm is cheap in comparison and it might add ~50mm2 to the IO-hub it would be well worth it for the tiny %-cost increase.




Well, Jaguar and 14h+16h carizzo-L were pretty low cost as far as VRM needs. In my opinion none of those cat APU's should have arrived as dual core APUs for consumers. And dual core APU salvage should've gone to embedded thin clients, or have had cores disabled for low end GPU. For consumer 3c/3t APU would have been the minimum for 14h, and near the minimum for 16h. For segmentation versus 4c/4t it would have been just right, especially with somewhat lower frequencies versus the well binned quadcores.

I think on my low end r7 250 card I see a 2 or 2+1 phase VRM.

How would you connect such a "GPU" to the motherboard, since the iGPU inside the APU connects directly to cpu ?

 

[hojnikb]

From my point of view the right solution to this is to a mobile CPU hooked to USB. Think a media stick with USB on one end and HDMI on the other. I really don't understand the need for a weak, low power GPU card (I mean, beyond a certain point). I can understand the desire for a cheap card, but ...

Thats a lot less elegant solution than just a regular type gpu. not to mention you'd be severly limited by power and thermals.

 

[amd6502]

How would you connect such a "GPU" to the motherboard, since the iGPU inside the APU connects directly to cpu ?

It seems like it should be technically possible but given the large pinout count and the somewhat more complex and costly VRM setup, who (besides AMD APU engineers) know if the whole solution could be practical and cost effective.

 

[cbn]

The whole point of such a card would be replacing outdated IGPs and add-in cards*. Not everybody can just wholesale replace PCs every 3rd year. Desktops have potential 10+ year lifespans these days. There isn't much of a market in the developed world, but I could see such would be useful elsewhere, where 2nd hand or even 3rd hand PCs are common.

Now, if such would be worth the resources to produce is a whole other question.

*With the side benefit of cutting power use massively in some cases. Older cards without proper power management are powerhogs at idle.

I think low end desktop cards with modern codec support is a really good thing.

However, one problem that I believe exists is conflict of interest with mid range APUs (e.g. R3 2200G).

Because AMD needs to sell R3 2200G there is less chance they will release low end cards. In fact, Nvidia has been the best source of low end cards for quite some time because of this.

P.S. What I am saying also has effect on AMD's releasing larger iGPU for the 2C/4T.

 

[LightningZ71]

What seems obvious to me, but maybe not to the world at large, is that there is a space for a modest low profile dGPU card that has an updated gpu in it with modern codecs and modest VRAM, but stays within the SFF PCI powerdraw specs and can route its output through the motherboard VGA/DP/HDMI connectors on systems that have iGPUs. These lower end CPUs shouldn't have any issues with 8x PCI bus congestion. If properly configured, they should also be compatible with DX12 mGPU setups in the few games that support that. It should help extend the life of sff computers immensely.

 

[VirtualLarry]

and can route its output through the motherboard VGA/DP/HDMI connectors on systems that have iGPUs.

How do you plan on doing that? If they could do that, they would have M.2 PCI-E GPUs for SFF rigs.

 

[amd6502]

Yes, but it goes through a decode process twice.

CPU core; ALU + AGU + FPU + FCU

FCU => Forwarder to Compute Unit;
The FCU's job is to make Graphic Core Next instructions from x86 FPU instructions. It is implied these x86 FPU ops would be a separate prefix than standard FPU ops.

In the more modern cases however, the FCU is integrated to the CPU front-end and the GCN back-end(TMUs/Graphic L1D/RBE) is integrated to the CPU back-end.

It is unlikely to get either options.

What do you mean these x86 FPU ops would be a separate prefix from standard FPU ops; and do you mean the micro ops or the original ops would have the different prefixes?

Why do you think a NG dozer is unlikely to make use of an FCU?

My vision for the NG dozer CCX would be 2+2 big-little configuration with the little ones running at half frequency, getting minimal share of the L2 cache, and tolerating high latencies. That means they would be an ideal candidate for x87 FCU substitution. You could skip the whole or at least much of the FPU logic with the FCU. So you have this pair of fast deep cores complemented by these two minimalist little cores running at half the frequencies.

 

[NTMBK]

What do you mean these x86 FPU ops would be a separate prefix from standard FPU ops; and do you mean the micro ops or the original ops would have the different prefixes?

Why do you think a NG dozer is unlikely to make use of an FCU?

My vision for the NG dozer CCX would be 2+2 big-little configuration with the little ones running at half frequency, getting minimal share of the L2 cache, and tolerating high latencies. That means they would be an ideal candidate for x87 FCU substitution. You could skip the whole or at least much of the FPU logic with the FCU. So you have this pair of fast deep cores complemented by these two minimalist little cores running at half the frequencies.

The whole point of the little core is to save power. You don't save power by going across the die and firing up the GPU every time you hit a floating point instruction.

 

[NostaSeronx]

Why do you think a NG dozer is unlikely to make use of an FCU?

The specific patent tree describes a different core.
https://patents.google.com/patent/US6801207B1/
https://patents.google.com/patent/US7023445B1/
https://patents.google.com/patent/US20090160863A1/
https://patents.google.com/patent/US7930519B2/

What do you mean these x86 FPU ops would be a separate prefix from standard FPU ops; and do you mean the micro ops or the original ops would have the different prefixes?

The ops are going to a separate co-processor. So, the FPU front-end does not want the instructions.

 

[ET]

How would you connect such a "GPU" to the motherboard, since the iGPU inside the APU connects directly to cpu ?

As I understand it, the idea is to have the iGPU on the I/O die. Which naturally connects to the motherboard (it's the I/O chiplet, after all; it does all the I/O work for the CPU chiplet).

Thats a lot less elegant solution than just a regular type gpu. not to mention you'd be severly limited by power and thermals.

Not sure why you think it's less elegant. In looks, perhaps (because it's not hidden inside a PC). But people are a lot more comfortable plugging an external device into a port than opening a PC, and an external device also fits a lot more form factors. The same display adaptor would work for a full case, a low profile one, a NUC style device or a laptop. To me, that's elegant.

And sure, it will be functionally limited. But the definition of the problem was something which provides up to date video standards at a low price for people who don't need 3D performance.

 

[amd6502]

Not sure why you think it's less elegant. In looks, perhaps (because it's not hidden inside a PC). But people are a lot more comfortable plugging an external device into a port than opening a PC, and an external device also fits a lot more form factors. The same display adaptor would work for a full case, a low profile one, a NUC style device or a laptop. To me, that's elegant.

I'm warming up to that idea now. You could actually supply a good amount of power with usb-c, and the video "card" could come with a mounting clip and attach right to the monitor. Perhaps even come with a mic, speaker, and a webcam for skype.

 

[hojnikb]

As I understand it, the idea is to have the iGPU on the I/O die. Which naturally connects to the motherboard (it's the I/O chiplet, after all; it does all the I/O work for the CPU chiplet).



Not sure why you think it's less elegant. In looks, perhaps (because it's not hidden inside a PC). But people are a lot more comfortable plugging an external device into a port than opening a PC, and an external device also fits a lot more form factors. The same display adaptor would work for a full case, a low profile one, a NUC style device or a laptop. To me, that's elegant.

And sure, it will be functionally limited. But the definition of the problem was something which provides up to date video standards at a low price for people who don't need 3D performance.

Again, USB itself is limited to 2,5W (or 4,5W for USB3) so designing a gpu around this is a challenge.
A re-purposed ARM SoC would be a better idea in that respect (low power and all the necessary codecs)

 

[Insert_Nickname]

And sure, it will be functionally limited. But the definition of the problem was something which provides up to date video standards at a low price for people who don't need 3D performance.

Again, USB itself is limited to 2,5W (or 4,5W for USB3) so designing a gpu around this is a challenge.
A re-purposed ARM SoC would be a better idea in that respect (low power and all the necessary codecs)

Perhaps it isn't a GPU we need so much as an external video decoder chip? Something like what was done for the earliest Atom platforms? Which had a separate decoder chip available.

With 10Gbit (and in future 20Gbit) USB3.1 external bandwidth shouldn't be an issue. I'm sure you could fit something like that in a 4.5W TDP. If you use a Type-C connector, you'd even have 3A (15W) to play with.

 

[hojnikb]

bandwidth not, but latency sure will be.

 

[ET]

Again, USB itself is limited to 2,5W (or 4,5W for USB3) so designing a gpu around this is a challenge.
A re-purposed ARM SoC would be a better idea in that respect (low power and all the necessary codecs)

That's precisely what I was talking about ("the right solution to this is to a mobile CPU hooked to USB"; yeah, can't write a complete sentence to save my life). There are cheap inexpensive chips with 4K H.264 and H.265 support, even VP9.

There are USB monitors, and they work quite well, so there's no particular reason why an intermediate step (a USB GPU hooked to a normal monitor) won't work.

 

[Abwx]

Oh the excitement.

Oh the trolling...

You dont get it, read the posts above, this APU is also for FM2+, at 33€ it replace advantageously a 2C Kaveri for whom use his PC for basic needs but want an updated UVD, if one wants more CPU grunt there s a 4C sibling at 50€...

https://www.mindfactory.de/product_info.php/AMD-A6-7480-2x-3-8GHz-So-FM2--BOX_1289834.html

For whom goes AM4 that s a good start in the waiting of lower priced Ryzens..

https://www.mindfactory.de/product_info.php/AMD-A6-9400-3-7GHZ-65W-2C_1296399.html